| _version_ | 1866902011171045376 |
|---|---|
| author | shrestha, Anuja |
| author_facet | shrestha, Anuja |
| contents | <p>This bachelor thesis focuses on the development of a deep learning-based system for automatic generation of radiology reports from chest X-ray images. Accurate diagnosis in healthcare relies heavily on medical imaging modalities such as X-rays, CT scans, and MRIs; however, a shortage of skilled radiologists causes delays in reporting. This project addresses the challenge by constructing models that use Convolutional Neural Networks (CNNs) for image feature extraction and Recurrent Neural Networks (RNNs) for generating textual reports. Utilizing the MIMIC-CXR dataset, several deep learning architectures, including VGG19, EfficientNetB0, DenseNet201, LSTM, BiLSTM, and GPT-2, were evaluated. The best results were achieved with the VGG19+LSTM model on the report “impression” section, with BLEU-1 and BLEU-2 scores of 0.6 and 0.56, respectively. The EfficientNetB0+Greedy Search model showed promising results on the “findings” section. This research demonstrates the potential of AI to assist radiologists by automating report generation, thereby improving diagnostic speed and patient care.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_15826800 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Visual Language Pre-Trained Model for Automated Medical Chest X-rays Report Generation shrestha, Anuja <p>This bachelor thesis focuses on the development of a deep learning-based system for automatic generation of radiology reports from chest X-ray images. Accurate diagnosis in healthcare relies heavily on medical imaging modalities such as X-rays, CT scans, and MRIs; however, a shortage of skilled radiologists causes delays in reporting. This project addresses the challenge by constructing models that use Convolutional Neural Networks (CNNs) for image feature extraction and Recurrent Neural Networks (RNNs) for generating textual reports. Utilizing the MIMIC-CXR dataset, several deep learning architectures, including VGG19, EfficientNetB0, DenseNet201, LSTM, BiLSTM, and GPT-2, were evaluated. The best results were achieved with the VGG19+LSTM model on the report “impression” section, with BLEU-1 and BLEU-2 scores of 0.6 and 0.56, respectively. The EfficientNetB0+Greedy Search model showed promising results on the “findings” section. This research demonstrates the potential of AI to assist radiologists by automating report generation, thereby improving diagnostic speed and patient care.</p> |
| title | Visual Language Pre-Trained Model for Automated Medical Chest X-rays Report Generation |
| url | https://doi.org/10.5281/zenodo.15826800 |